The Transmission Control Protocol/Internet Protocol (TCP/IP) protocol suite allows computers of all sizes, from many different computer vendors, running totally different operating systems, to communicate with each other. Stevens, W. Richard, 1994, TCP/IP Illustrated, Volume 1, which is incorporated herein by reference, describes general aspects of the TCP/IP protocol suite. Request For Comment (RFC) 793, also incorporated by reference, is considered the main TCP specification. Additional RFCs that describe TCP and IP are published for informational purposes. The RFCs are provided by the Internet Engineering Task Force (IETF) at www.ietf.org.
FIG. 1 is a block diagram illustrating a networked system consisting of a pair of computer systems (nodes) 12 and 14 executing the TCP/IP protocol suite. In FIG. 1, nodes 12 and 14 communicate with each other by transmitting frames of data via router 16 and Ethernet communication links 20 and 22.
FIG. 2 is a block diagram illustrating relevant components the TCP/IP protocol suite. More particularly, nodes 12 and 14 are shown having four communication layers 30-36 and 40-46, respectively. Layers 30-36 and 40-46 take form in software executing on one or more processors in nodes 12 and 14, respectively. Layers 32-36 and 42-46 are implemented in operating systems of nodes 12 and 14, respectively, while layers 30 and 40 take form in any one of many user applications including File Transfer Protocol (FTP), Simple Mail Transfer Protocol (SMTP), telnet for remote login, etc.
Layers 32 and 42 of nodes 12 and 14, respectively, are commonly referred to as the transport layers. There are several distinct transport layers including TCP and User Datagram Protocol (UDP). For purposes of explanation, layers 32 and 42 will take form in either the TCP or UDP transport layers. Layers 34 and 44 are commonly referred to as the network layers. A network layer may take many forms such as Internet Control Message Protocol (ICMP) or IP. For purposes of explanation, FIG. 2 will be described with layers 34 and 44 taking form in the well-known IP network layer. Lastly, layers 36 and 46 are commonly referred to as the link or network interface layers. This layer handles the details of physically interfacing with a communication link. For purposes of explanation, link layers 36 and 46 will take form in Ethernet link layers for interfacing with Ethernet communication links 20 and 22, respectively.
Most networked systems are designed such that at least one node is a client to a server node. In FIG. 2, node 12 is presented as a client to server node 14. In this configuration, server application 40 provides some type of service (e.g., SMTP) in response to a request from client application 30. When client application 30 communicates with server application 40, including a request for service, client application 30 sends data down through layers 32-36 until the data is sent as a stream of bits to node 14 via router 16 and communication links 20 and 22. The data received by node 14 is sent up through layers 46-42 until it reaches server application 40.
As data moves down layers 30-36 of node 12, each of the layers 30-36 adds headers (and sometimes trailers) containing communication information. FIG. 3 illustrates relevant aspects of this process. FIG. 3 shows data 50 generated by application layer 30. An application header 52 is concatenated to data 50 by application layer 30, the result of which is provided to TCP/UDP layer 32 as application data 54. TCP/UDP layer 32, in turn, concatenates a TCP or UDP header to application data 54 received from application layer 30. For purposes of explanation, TCP/UDP layer 32 concatenates a TCP header 56 to application data 54, the result of which is provided to IP layer 34. The unit of data that the TCP layer 32 sends to IP layer 34 is called an IP payload. IP layer 34 concatenates an IP header 60 to the IP payload it receives, the result of which is provided to Ethernet link 36. Generally, the unit of data that layer 34 sends to layer 36 is referred to as an IP datagram (datagram) or packet. As will be more fully described below, a datagram can be fragmented and sent to Ethernet layer 36 as datagram fragments. Ethernet layer 36 appends an Ethernet header 62 and Ethernet trailer 64 to the datagram or datagram fragments received from IP layer 34, the result of which is referred to as an Ethernet frame.
The Ethernet frame is transmitted to node 14 as a stream of bits via router 16 and Ethernet communication links 20 and 22 (FIG. 2). When the Ethernet frame is received at destination node 14, the frame moves up through the protocol layers 46-42, and all headers (and trailers) are removed by the appropriate protocol layer until the original data 50 is provided to server application 40. Each layer looks at certain identifiers in its corresponding header to determine which succeeding layer is to receive the data. This is called demultiplexing. For example, Ethernet layer 46 routes the datagram or datagram fragment of the frame it receives to IP layer 44 after Ethernet layer 46 strips off the Ethernet header and trailer 62 and 64, respectively.
FIG. 4 shows an exemplary Ethernet frame in greater detail. The Ethernet frame includes header 62, datagram 64, and trailer 66. The header 62 consists of a 6-byte Ethernet destination address, a 6-byte Ethernet source address, and a 2-byte type code. The Ethernet trailer 66 consists of a Cyclic Redundancy Check (CRC) field. CRC field is used to detect errors in the rest of the frame. Figures in this specification, including FIG. 4, show an empty 2-byte field in the Ethernet frame header adjacent the 2-byte type code. This empty field contains no data. This empty field is simply provided in the Figures so that datagram 64 can be positioned to begin at the first bit of the first line after the header.
FIG. 5 illustrates an exemplary IP datagram having an IP header 80 and IP payload (e.g., TCP segment) 82. The IP header shown in FIG. 5 includes many fields each of which stores a value. The 4-bit header length field defines the number of 32-bit words in the IP header, including any option fields. The 16-bit total length field defines the total length of the IP datagram in bytes. The header length subtracted from the total length can be used to determine the IP payload (e.g., TCP segment) length. The 8-bit protocol field defines the transport layer (e.g., TCP or UDP) that is to receive the IP payload of the datagram.
Normally, an upper limit is imposed on the size of the frame that can be transmitted between nodes 12 and 14. Many IP datagrams are fragmented to meet this limit. Thus, an IP datagram may be fragmented into IP datagram fragments, each of which is provided to the Ethernet layer for transmission in a separate frame. When an IP datagram is fragmented it is not reassembled until it reaches its final destination. The IP layer at the destination performs the reassembly. The 16-bit identification field shown in FIG. 5 contains a unique value for each IP datagram that the sender transmits. This number is copied into each fragment of a particular datagram. The 13-bit fragment offset field of the header shown in FIG. 5, contains the offset (in 8 byte units), of the fragment from the beginning of the original IP datagram. When an IP datagram is fragmented, the total length field of each fragment is changed to be the size of that fragment.
FIGS. 6 and 7 illustrate UDP and TCP segments, respectively. In FIG. 7, the TCP segment (IP payload) includes a TCP header 90 and application data. TCP header 90 includes several fields each containing a value. The 4-bit header length defines the length of the TCP header in 32-bit words. When a new connection is sought to be established between nodes 12 and 14 for the purposes of, for example, file transfer, the 1-bit SYN flag of the TCP header 90 is turned on or set to binary 1. TCP headers as well UDP headers include 16-bit port number fields. Servers, such as node 14, are normally known by their well-known port number. For example, every TCP/IP implementation that provides an FTP server provides that service on TCP port number 21. Telnet servers are commonly on TCP port number 23.
Unauthorized access (i.e., hacking) of computer systems, such as node 14 of FIG. 2, continues to be an ongoing problem. Filters are often provided in routers, such as router 16, for preventing unauthorized access of computer systems. Filters, oftentimes referred to as firewalls, may take form in hardware and/or software executing on one or more processors. FIG. 2 shows router 16 having a filter 100. Filter 100 checks frames (or the IP datagrams thereof) received by router 16 to ensure that the frames are not designed to initiate an unauthorized operation at node 14. If a frame is deemed by filter 100 to be part of an authorized operation, the frame is allowed to pass through the filter. If the frame is deemed by the filter to be part of an unauthorized operation, the frame is dropped by filter 100 so that the frame cannot reach its final destination (i.e., node 14).
Filter 100 performs numerous checks on frames it receives. For example, filter 100 compares the destination port number and SYN flag values of all frames it receives. If the destination port number and SYN flag values of a received frame equal 21 and binary 1, respectively, the received frame is configured to initiate an unauthorized file transfer protocol (FTP) and will be dropped by filter 100. FIG. 8 shows a frame 102 having a TCP header 90 with a destination port number and a SYN flag set to 21 and binary 1, respectively. Frame 102 is configured to initiate a file transfer at node 14. Filter 100 will drop frame 102.
As noted above, IP datagrams can be fragmented to meet the limit on the size of frames that can be transmitted between nodes 12 and 15. Fragmented IP datagrams are not reassembled until they reach their final destination, e.g., node 14. Thus, Ethernet frames containing datagram fragments can pass through router 16, and filter 100 thereof, before reaching destination node 14. Fragmentation can be used to disguise frames from filter 100. One technique for disguising frames using fragmentation is often referred to as a “tiny fragment attack.” For example, FIGS. 9 and 10 illustrate frames containing datagram fragments which, when reassembled at node 14, can initiate an otherwise unauthorized file transfer operation. More particularly, FIG. 9 shows a frame 104 consisting of an Ethernet header 62, datagram fragment 64, and CRC 66. Datagram fragment 64 of frame 104 includes full IP and TCP headers. Frame 106 shown in FIG. 10 also includes an Ethernet header 62, datagram fragment 64 and CRC 66. Datagram fragment 64 includes an IP header 80 with fragment offset set to 0 and a less than full TCP header 90. As such, datagram fragment 64 of frame 106 is referred to as a tiny fragment. As will be more fully described below, filter 100 will pass frames 104 and 106 to node 14. FIG. 11 shows the datagram payload which results when node 14 reassembles the payloads of the datagram fragments of frames 104 and 106. The datagram payload of FIG. 11 initiates an otherwise unauthorized FTP at node 14.
Frames 104 and 106 shown in FIGS. 9 and 10, respectively, are transmitted to router 16 in sequence. As noted above, one of the checks performed by filter 100 is to compare the destination port number and SYN flag values to 21 and binary 1, respectively, of received frames. In frame 104 of FIG. 9, the destination port number in TCP header 90 is set to 23 while the SYN flag is set to binary 1. With these destination port number and SYN flag values, frame 104 passes the filter check mentioned above, and frame 104 is forwarded to node 14. Frame 106 shown in FIG. 10 includes only a source port number, a destination port number, and a sequence number. Filter 100 will pass frame 106 since frame 106 lacks a destination port number equal to 21 and SYN flag equal to binary 1. It is noted that the fragment offset is set to 0 within frame 106. As such, when the datagram payload fragments contained in frames 104 and 106 are reassembled at node 14, the source port number, destination port number, and sequence number of the TCP header 90 in FIG. 9 will be overwritten with the source port number, destination port number, and sequence number, respectively, of the TCP header 90 shown in FIG. 10, resulting in the datagram shown in FIG. 11.
The tiny fragment attack described above can be averted. More particularly, RFC 3128 describes an algorithm which may be used within filter 100 to avert the tiny fragment attack described above. The RFC 3128 algorithm provides:
if (Protocol = TCP)  if(Fragment Offset = 0)Check Length 1 ≧ 16 bytes  elseCheck Fragment Offset ≧ 2,where Length 1 is calculated by filter 100 for each frame according to the following equation:Length 1 (in bytes)=(Total Length Value in IP Header of the Received Frame)−((IP Header Length Value in IP Header of the Received Frame)×4).
In accordance with the RFC 3128 algorithm above, filter 100 will drop any frame it receives if the protocol field of the IP header is set to TCP, the fragment offset of the IP header is set to 0, and the calculated Length 1 is less than 16 bytes. Additionally, filter 100 will drop any received frame if the fragment offset of the IP header is set to 1. Frame 104 shown in FIG. 9 will pass both checks performed by filter 100 executing the RFC 3128 algorithm above. However, frame 106 shown in FIG. 10 will be dropped by filter 100 since the calculated Length 1 of frame 106 is 8 bytes.
The RFC 3128 algorithm set forth above is incapable of averting certain variations of the tiny fragment attack described above. To illustrate, FIG. 12 shows a frame 108 having an Ethernet header 62, a datagram 64, and CRC 66. As can be seen in FIG. 12, TCP header 90 of frame 108, like TCP header 90 shown in FIG. 10, is less than a complete or full TCP header. Note also that the value of the total length field of the IP header is 36 bytes in frame 108 even though actual total length of the datagram 64 is 28 bytes. Frame 108 may pass RFC 3128 algorithm above since the calculated Length 1 is 16 bytes. Unfortunately, if the frame 108 passes filter 100 and datagram 64 of frame 108 combines with random data in memory of node 14, the result may produce an unauthorized operation such as file transfer protocol.